Merge branch 'master' of git://git.denx.de/u-boot-arm

[oweals/u-boot.git] / cpu / blackfin / start.S

diff --git a/cpu/blackfin/start.S b/cpu/blackfin/start.S
index 30212e92818794e3b125d7b3cba2cdaa7fd7f8ec..6c8def48162fef0cd0ec23cfa9bbbf820b5610ae 100644 (file)
--- a/cpu/blackfin/start.S
+++ b/cpu/blackfin/start.S
@@ -1,7 +1,7 @@
 /*
  * U-boot - start.S Startup file for Blackfin u-boot
  *
- * Copyright (c) 2005-2007 Analog Devices Inc.
+ * Copyright (c) 2005-2008 Analog Devices Inc.
  *
  * This file is based on head.S
  * Copyright (c) 2003  Metrowerks/Motorola
@@ -49,8 +49,8 @@
 ENTRY(_start)
 
        /* Set our initial stack to L1 scratch space */
-       sp.l = LO(L1_SRAM_SCRATCH + L1_SRAM_SCRATCH_SIZE);
-       sp.h = HI(L1_SRAM_SCRATCH + L1_SRAM_SCRATCH_SIZE);
+       sp.l = LO(L1_SRAM_SCRATCH_END - 20);
+       sp.h = HI(L1_SRAM_SCRATCH_END - 20);
 
 #ifdef CONFIG_HW_WATCHDOG
 # ifndef CONFIG_HW_WATCHDOG_TIMEOUT_START
@@ -75,7 +75,7 @@ ENTRY(_start)
 
        serial_early_puts("Init Registers");
 
-       /* Disable nested interrupts and enable CYCLES for udelay() */
+       /* Disable self-nested interrupts and enable CYCLES for udelay() */
        R0 = CCEN | 0x30;
        SYSCFG = R0;
 
@@ -125,8 +125,11 @@ ENTRY(_start)
         */
        r6 = 1 (x);
 
-       /* Relocate from wherever are (FLASH/RAM/etc...) to the
-        * hardcoded monitor location in the end of RAM.
+       /* Relocate from wherever we are (FLASH/RAM/etc...) to the hardcoded
+        * monitor location in the end of RAM.  We know that memcpy() only
+        * uses registers, so it is safe to call here.  Note that this only
+        * copies to external memory ... we do not start executing out of
+        * it yet (see "lower to 15" below).
         */
        serial_early_puts("Relocate");
        call _get_pc;
@@ -135,27 +138,16 @@ ENTRY(_start)
        r2.h = .Loffset;
        r3.l = _start;
        r3.h = _start;
-       r1 = r2 - r3;
-
-       r0 = r0 - r1;
-
-       cc = r0 == r3;
+       r2 = r2 - r3;
+       r1 = r0 - r2;
+       cc = r1 == r3;
        if cc jump .Lnorelocate;
-
        r6 = 0 (x);
-       p1 = r0;
 
-       p2.l = LO(CFG_MONITOR_BASE);
-       p2.h = HI(CFG_MONITOR_BASE);
-
-       p3 = 0x04;
-       p4.l = LO(CFG_MONITOR_BASE + CFG_MONITOR_LEN);
-       p4.h = HI(CFG_MONITOR_BASE + CFG_MONITOR_LEN);
-.Lloop1:
-       r1 = [p1 ++ p3];
-       [p2 ++ p3] = r1;
-       cc=p2==p4;
-       if !cc jump .Lloop1;
+       r0 = r3;
+       r2.l = LO(CONFIG_SYS_MONITOR_LEN);
+       r2.h = HI(CONFIG_SYS_MONITOR_LEN);
+       call _memcpy_ASM;
 
        /* Initialize BSS section ... we know that memset() does not
         * use the BSS, so it is safe to call here.  The bootrom LDR
@@ -173,37 +165,41 @@ ENTRY(_start)
 .Lnorelocate:
 
        /* Setup the actual stack in external memory */
-       r0.h = HI(CONFIG_STACKBASE);
-       r0.l = LO(CONFIG_STACKBASE);
-       sp = r0;
+       sp.h = HI(CONFIG_STACKBASE);
+       sp.l = LO(CONFIG_STACKBASE);
        fp = sp;
 
        /* Now lower ourselves from the highest interrupt level to
         * the lowest.  We do this by masking all interrupts but 15,
-        * setting the 15 handler to "board_init_f", raising the 15
+        * setting the 15 handler to ".Lenable_nested", raising the 15
         * interrupt, and then returning from the highest interrupt
         * level to the dummy "jump" until the interrupt controller
-        * services the pending 15 interrupt.
+        * services the pending 15 interrupt.  If executing out of
+        * flash, these steps also changes the code flow from flash
+        * to external memory.
         */
        serial_early_puts("Lower to 15");
        r0 = r7;
        r1 = r6;
        p0.l = LO(EVT15);
        p0.h = HI(EVT15);
-       p1.l = _cpu_init_f;
-       p1.h = _cpu_init_f;
+       p1.l = .Lenable_nested;
+       p1.h = .Lenable_nested;
        [p0] = p1;
-       p2.l = LO(IMASK);
-       p2.h = HI(IMASK);
-       p3.l = LO(EVT_IVG15);
-       p3.h = HI(EVT_IVG15);
-       [p2] = p3;
+       r7 = EVT_IVG15 (z);
+       sti r7;
        raise 15;
        p4.l = .LWAIT_HERE;
        p4.h = .LWAIT_HERE;
        reti = p4;
        rti;
 
+       /* Enable nested interrupts before continuing with cpu init */
+.Lenable_nested:
+       cli r7;
+       [--sp] = reti;
+       jump.l _cpu_init_f;
+
 .LWAIT_HERE:
        jump .LWAIT_HERE;
 ENDPROC(_start)